XFS文件系统删除重复文件后Inode总数变动问题咨询

Hey there, looking at your df -i outputs where the total inode count jumps from ~46 million to 114 million then down to 94 million (while used inodes stay roughly the same), this is tied to how XFS manages inodes differently from more familiar filesystems like ext4. Let's break down the reasons and troubleshooting steps:

你提供的df -i输出显示inode总数波动明显：

my-linux:/content# df -i .
Filesystem            Inodes   IUsed   IFree IUse% Mounted on
/dev/md3             46295616 43991622 2303994   96% /content

my-linux:/content# df -i .
Filesystem            Inodes   IUsed   IFree IUse% Mounted on
/dev/md3             114738816 43959820 70778996   39% /content

my-linux:/content# df -i .
Filesystem            Inodes   IUsed   IFree IUse% Mounted on
/dev/md3             94207328 43960396 50246932   47% /content

核心原因：XFS的动态inode分配机制 #

Unlike ext4 (which fixes the total number of inodes at format time), XFS uses dynamic inode allocation by default. It doesn't pre-allocate all possible inodes upfront—instead, it creates new inodes from free disk space as needed, up to a configured limit (usually 25% of total disk space, adjustable at format time). The Inodes value in df -i shows the current number of allocated inodes, not a fixed maximum.

That said, the fluctuation (going up then down) isn't standard behavior for basic dynamic allocation, so let's dig into specific checks:

1. Verify XFS filesystem configuration #

First, check your XFS setup with xfs_info /dev/md3. Look for these key parameters:

• icount: The initial number of inodes pre-allocated at format time
• imaxpct: The maximum percentage of disk space allowed for inode storage (default 25%)
• inodesize: The size of each inode (affects how many fit in a given space)

If imaxpct is set high, XFS will be more aggressive about creating inodes. If you have plenty of free disk space after deleting duplicates, the system might have expanded the inode pool, then possibly trimmed it if some inodes were marked as unused (though XFS doesn't usually shrink the inode pool automatically—this points to other factors).

2. Check for snapshots or cloned volumes #

XFS snapshots (or underlying LVM snapshots) preserve the state of inodes from the original filesystem, even after files are deleted. If your system runs automatic snapshot backups, creating or deleting snapshots will alter the total inode count reported by df -i.

• For XFS native snapshots: Run xfs_admin -l /dev/md3 to list existing snapshots
• For LVM snapshots: Use lvdisplay to check for snapshot logical volumes

3. Look for background filesystem maintenance processes #

System tools that maintain XFS can trigger inode count changes:

• xfs_scrub: An online checker/repair tool that might re-scan or adjust inode metadata
• xfs_fsr: A defragmentation tool that can rearrange inodes during file optimization
• Check running processes with ps aux | grep xfs to see if any of these are active

4. Validate df -i accuracy with direct superblock queries #

Sometimes df relies on cached metadata, which can cause temporary inaccuracies. Try these steps:

1. Run sync to force all pending metadata writes to disk
2. Re-run df -i . to see if the count stabilizes
3. For a definitive count, query the XFS superblock directly with:

   xfs_db -c "sb 0" -c "print icount" /dev/md3
   

   This reads the inode count straight from the filesystem's superblock, bypassing any df caching.

5. Check for hidden or orphaned inodes #

Even after deleting duplicates, some inodes might still be tied to:

• Hidden files/directories (use find /content -type f -name ".*" to locate them)
• Orphaned files: Files deleted but still held open by a process (find these with lsof | grep deleted)
  These can keep inodes in use, while XFS might allocate new inodes for temporary system operations, leading to apparent fluctuations.

内容的提问来源于stack exchange，提问作者Jae Sung Seo